543
ISSN 1472-2739 (on-line) 1472-2747 (printed)
Algebraic & Geometric Topology
Volume 4 (2004) 543–569
Published: 15 July 2004
ATG
Automorphisms of free groups with boundaries
Craig A. Jensen
Nathalie Wahl
Abstract The automorphisms of free groups with boundaries form a family of groups An,k closely related to mapping class groups, with the standard automorphisms of free groups as An,0 and (essentially) the symmetric
automorphisms of free groups as A0,k . We construct a contractible space
Ln,k on which An,k acts with finite stabilizers and finite quotient space and
deduce a range for the virtual cohomological dimension of An,k . We also
give a presentation of the groups and calculate their first homology group.
AMS Classification 20F65, 20F28; 20F05
Keywords Automorphism groups, classifying spaces.
1
Introduction
The group An,k of automorphisms of free groups with k boundaries was introduced in [29] to show that the natural map from the stable mapping class group
of surfaces to the stable automorphism group of free groups gives an infinite
loop map on the classifying spaces of the groups after plus construction. The
“boundaries” make it possible to define appropriate gluing operations.
There are several ways to define these groups.
Geometrically, An,k is the group of components of the homotopy equivalences
of a certain genus n + k graph Gn,k (see Figure 1) fixing k circles in the
graph. The mapping class group Γg,k+1 of a surface of genus g with k + 1
boundary components is in fact the subgroup of A2g,k fixing one more circle
(see Section 2). An alternative geometric description is given in [11] in terms
of the mapping class groups of a 3-dimensional manifold.
Algebraically, we first define a related group Akn , which corresponds to mapping
class groups with punctures rather than boundaries. Let Fn+k denote the free
group on the n + k generators x1 , . . . , xn , y1 , . . . , yk , and let Aut(Fn+k ) be its
automorphism group. Then Akn := {φ ∈ Aut(Fn+k ) | φ(yi ) ∼ yi for 1 ≤ i ≤ k},
c Geometry & Topology Publications
544
Craig A. Jensen and Nathalie Wahl
where ∼ means ‘conjugate to’. The groups Akn are thus natural intermediates between A0n = Aut(Fn ) and Ak0 which is known as the pure symmetric
automorphism group of Fk . The group An,k is a central extension
1 → Zk → An,k → Akn → 1
(see Section 2 for details).
The symmetric automorphisms of free groups have been studied extensively,
in particular for their relation to automorphism groups of free products (see
for example [1, 4, 14, 25, 26]) and for their relation to motion groups of strings
[3, 6, 7, 26]; they give a generalization of the braid groups, which is the mapping
class group of a genus 0 punctured surface. Our groups are the higher genus
analogues, coming from surfaces with punctures for Akn or with boundaries for
An,k .
In [18, Section 4], Levitt defines a class of groups, which he calls relative automorphism groups, in order to study automorphism groups of hyperbolic groups
and of graphs of groups. The groups An,k are examples of relative automorphism groups.
In this paper, we construct a contractible space Ln,k of dimension 2n + 3k − 2
(provided n ≥ 1; if n = 0 it is of dimension 2k−1) on which the group An,k acts
with finite stabilizers and such that the quotient space is a finite complex. The
space Ln,k is an analogue of the spine of Auter space Xn . Recall that Xn is a
space of graphs of genus n marked by a rose ∨n S 1 . In Ln,k , we consider graphs
of genus n + k marked by the graph Gn,k of Figure 1. We also construct an
analogous space Lkn for the group Akn so that there is a sequence of contractible
spaces
Fn,k → Ln,k → Lkn
corresponding to the group extension Zk → An,k → Akn . The fiber Fn,k is
homeomorphic to Rk with the standard Zk -action.
We use Ln,k to give the following range for the virtual cohomological dimension
of An,k : 2n + 2k − 1 ≤ vcd(An,k ) ≤ 2n + 3k − 2 when k ≥ 1 (Theorem 3.5).
By the work of McCool [24], we know that Akn is finitely presented. We use
his work to produce a presentation for Akn and An,k (Theorems 7.1 and 7.2).
We recover the presentation he gave for Ak0 in [25]. Theorem 7.4 says that
H1 (An,k ) = Z2 when n > 2. This agrees with the recently proved conjecture
that the homology Hi (An,k ) is independent of n and k for n >> i [11].
To show that Ln,k is contractible, we show that it is isomorphic to a subcomplex
Θ
of Xn+3k
, the set of fixed points of Xn+3k under the action of a certain finite
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
545
subgroup Θ of Aut(Fn+3k ). Although An,k is not a subgroup of Aut(Fn+k ), it
Θ
is a subgroup of the normalizer of Θ in Aut(Fn+3k ). Hence An,k acts on Xn+3k
,
which is known to be contractible [13]. The space Ln+k can be identified with
Θ
the essential realization of an initial segment of the essential realization of Xn+3k
under a certain norm, in the language of [17]. Our Theorem 4.1 says that the
initial segments of the essential realization of XnG are contractible for any finite
group G 6 Aut(Fn ).
Our motivation is the study of the stable automorphism group Aut∞ , the
colimit of the inclusions Aut(Fn ) ,→ Aut(Fn+1 ). The homology and homotopy
type of the classifying space of the stable mapping class group Γ∞ has been
determined recently [19, 20, 28], making use of the mapping class groups Γg,n
of surfaces with any number of boundary components, and the fact that the
homology groups H∗ (Γg,n ) are independent of g and n when g >> ∗ [8].
Little is known about BAut∞ . It was shown recently that the homology of
An,k is also independent of the number of boundaries in a stable range [11].
This means in particular that we can work with An,k instead of Aut(Fn ) if we
are only interested in stable results. The boundaries of An,k allow us to define
gluing operations analogous to the very useful gluing operations for the mapping
class groups. These give a wealth of new operations on the automorphism of
free groups and are used in [29] to show that the map on classifying spaces
+
BΓ+
∞ → BAut∞ is an infinite loop map after Quillen’s plus-construction.
The paper is organized as follows. Section 2 recalls the definition of the groups
An,k and describes their relationship to mapping class groups of surfaces. Section 3 defines the space Ln,k and calculates a range for the virtual cohomological
dimension of An,k . In Section 4, we prove the contractibility of initial segments
Θ
in a general setting. Section 5 shows that Ln,k is an initial segment of Xn+3k
.
k
In Section 6, we exhibit the sequence Fn,k → Ln,k → Ln . Finally, Section 7
gives the presentations and calculates H1 .
Acknowledgment
The authors would like to thank Benson Farb for helpful conversations and
the referee for interesting suggestions. The work was conducted while the first
author was partially supported by Louisiana Board of Regents Research Competitiveness Subprogram Contract LEQSF-RD-A-39 and second was supported
by a Marie Curie Fellowship of the European Community under contract number HPMF-CT-2002-01925.
Algebraic & Geometric Topology, Volume 4 (2004)
546
2
Craig A. Jensen and Nathalie Wahl
The groups A n,k
Let Gn,k be the graph given in Figure 1, obtained from a wedge of n circles by
attaching k circles C1 , . . . , Ck on stems. We think of the circles (together with
the basepoint) as the boundary of the graph. The basepoint ∗ of Gn,k is the
central vertex. Define
An,k := π0 Htpy∗ (Gn,k ; ∂),
where Htpy∗ (Gn,k ; ∂) is the space of (basepointed) homotopy equivalences of
Gn,k fixing the circles Cj point-wise. By [9, Prop. 0.19], homotopy equivalences
fixing the circles are actually homotopy equivalences relative to the circles. This
shows in particular that An,k is a group.
y2
y1
yk
x1
xn
x2
Figure 1: Gn,k
Labeling the first n petals of Gn,k with x1 , . . . , xn and the k circles with
their stems y1 , . . . , yk induces an identification of π1 (Gn+k ) with the free group
Fn+k := hx1 , . . . , xn , y1 , . . . , yk i. This leads to the following algebraic description of An,k : Define a map α : (Fn+k )n+k → Hom(Fn+k , Fn+k ) by
αhν, wi[xi ] = νi
1≤i≤n
αhν, wi[yj ] = wj−1 yj wj 1 ≤ j ≤ k
where hν, wi := (ν1 , . . . , νn , w1 , . . . , wk ) ∈ (Fn+k )n+k .
Proposition 2.1 There is a group isomorphism
An,k ∼
= {hν, wi ∈ (Fn+k )n+k αhν, wi ∈ Aut(Fn+k )
with the multiplication on the right hand side defined by
hν, wi.hν 0 , w0 i = h αhν 0 , w0 i[ν] , w0 . αhν 0 , w0 i[w] i,
where we apply αhν 0 , w0 i component-wise, and the unit is (x1 , . . . , xn , 1, . . . , 1).
Algebraic & Geometric Topology, Volume 4 (2004)
547
Automorphisms of free groups with boundaries
Proof We use the labeling of Gn,k by xi ’s and yj ’s as above (with a fixed
orientation) to get an identification π1 (G) ∼
= Fn+k . The “lollies” yj are made
out of a stem, denoted lj , from the basepoint to the circle, denoted Cj .
For any f ∈ Htpy∗ (G; ∂), we want to define an element hν, wi as above which
depends only on the homotopy class of f . Set νi := [f (xi )] and wj := [lj .f (lj )]
in π1 (G), where lj means the path lj taken in the reverse direction. Now
αhν, wi is the automorphism induced by f on π1 (G). Indeed, [f (xi )] = νi and
[f (yj )] = [f (lj ).f (Cj ).f (lj )] which is equal to wj−1 yj wj in π1 (G). So hν, wi is
an element of α−1 (Aut(Fn+k )) as required.
Conversely, we want to show that any such hν, wi defines an element [f ] ∈
π0 Htpy∗ (G; ∂). Define f piecewise by f (xi ) = νi , f (lj ) = wj−1 .lj and f (Cj ) =
Cj . It is a continuous map, and a homotopy equivalence as it acts on π1 (G)
via αhν, wi.
The two maps are homomorphisms and inverse of one another. Hence the two
groups are isomorphic.
The map α : An,k → Aut(Fn+k ) is a homomorphism (where the multiplication
of automorphisms φ.ψ is defined by first applying φ and then ψ .) Geometrically, it is the map that forgets that the circles were fixed. It is not injective.
There is in fact a short exact sequence
i
α
1 → Zk → An,k → Akn → 1
where Akn ∼
= {φ ∈ Aut(Fn+k ) | φ(yj ) ∼ yj for 1 ≤ j ≤ k} and where
i(z1 , . . . , zk ) = (x1 , . . . , xn , y1z1 , . . . , ykzk ). Note that the group Akn can be described geometrically as π0 Htpy∗ (Gn,k ; [∂]), where Htpy∗ (Gn,k ; [∂]) is now the
space of homotopy equivalences of Gn,k fixing the circles up to a rotation.
The group Zk maps to the center of An,k . These k additional generators in
An,k correspond to “conjugating yi by itself”. Geometrically, these are Dehn
twists along the ith boundary component in the corresponding mapping class
group Γg,k+1 .
We will also consider the semidirect product AΣ
n,k = Σk n An,k . Geometrically,
Σ
Σ
AΣ
:=
π
Htpy
(G
;
∂)
with
Htpy
(G
;
∂)
the space of homotopy equiva0
n,k
n,k
∗
∗
n,k
lences fixing the cycles Cj up to a permutation. Algebraically, this means that
the yi ’s can be permuted.
The groups An,k and Akn are closely related to mapping class groups when n
is even. Indeed, note that the graph G2g,k is homotopy equivalent to a surface
Sg,k+1 of genus g with k + 1 boundary circles and An,k can also be defined as
Algebraic & Geometric Topology, Volume 4 (2004)
548
Craig A. Jensen and Nathalie Wahl
the group of components of the homotopy equivalences of Sg,k+1 fixing the first
k boundary components point-wise and a point on the last one.
Let Γg,k := π0 Diff + (Sg,k+1 ; ∂) denote the mapping class group of Sg,k+1 with
the boundaries point-wise fixed and let Γk+1
be the mapping class group of
g
Sg,k+1 where the boundaries are fixed, but no longer point-wise. Fricke (1897)
and Magnus (1934) for g = 0, 1 and Nielsen (1927) for g ≥ 2 proved that Γk+1
g
is the subgroup of Out(F2g+k ) fixing the cyclic words defined by the boundaries
[21, p175].
Note that the inner automorphisms of Fn+k form a normal subgroup of Akn . Let
Okn := Akn / Inn(Fn+k ). So Okn = {φ ∈ Out(Fn+k ) | φ(yj ) ∼ yj for 1 ≤ j ≤ k}.
It follows that Γk+1
is the subgroup of Okn fixing the conjugacy class of the
g
last boundary component, that is the word c := [x1 , x2 ] . . . [x2g−1 , x2g ] y1 . . . yk .
One actually has the following isomorphism of short exact sequences:
1
Zk+1
1
Γg,k+1
Γk+1
g
3
(z, φ) ∈ Zk × Inn(F2g+k ) φ(c) = c
∼
=
/
hν, wi ∈ A2g,k αhν, wi[c] = c
∼
=
/
φ ∈ Ok2g φ(c) ∼ c
∼
=
/
1
1
(n, k)-graphs
By a graph, we mean a pointed, connected, one-dimensional CW-complex. Let
S 1 be the circle with basepoint denoted by ◦.
Definition 3.1 An (n, k)-graph (Γ, l) is a graph Γ of genus (n + k) with no
separating edges, equipped with embeddings lj : S 1 → Γ for j = 1, . . . , k such
that
(1) For every j = 1, . . . , k , the point ◦j := lj (◦) is a vertex of Γ.
Algebraic & Geometric Topology, Volume 4 (2004)
549
Automorphisms of free groups with boundaries
(2) The vertices of Γ have valence at least 2 and if a vertex v has valence 2,
then v = ◦j for some j or v is the basepoint ∗.
(3) Any two Sj := lj (S 1 ) meet in at most a point and the dual graph of the
Sj ’s is a forest.
We call the embedded circle (Sj , ◦j ) a cycle. In this way, an (n, k)-graph is a
graph of genus n + k with k cycles. (See Figure 2 for an example.)
S3
S1
◦1 = ◦3
◦2
S2
∗
Figure 2: Example of a (2, 3)-graph
Let (Gn,k , κ) be the generalized (n, k)-graph, where Gn,k was defined in Section 2 (see Figure 1) and κj : S 1 → Gn,k identifies S 1 with the j th circle
Cj . Note that Gn,k has k separating edges, and hence this is not an actual
(n, k)-graph.
Definition 3.2 A marked (n, k)-graph (Γ, φ) is a graph Γ equipped with a
pointed homotopy equivalence φ : Gn,k → Γ such that (Γ, φ ◦ κ) is a (n, k)graph.
Two marked (n, k)-graphs (Γ1 , φ1 ) and (Γ2 , φ2 ) are equivalent if there is a
graph isomorphism ψ : Γ1 → Γ2 such that ψ ◦ φ1 (Cj ) = φ2 (Cj ) and ψ ◦ φ1 ' φ2
(rel. to Cj for all j .)
The intuition (which is made precise in the proof of Theorem 5.3) is that condition (3) in Definition 3.1 prevents the existence of Whitehead moves (blowing
an edge and then collapsing it) which would take a ‘good marking’ to a ‘bad
marking’.
For an (n, k)-graph (Γ, l), let Γ̂ be the graph of genus n obtained from Γ by
collapsing all the edges in the cycles S1 , . . . , Sk . Let e be an edge of Γ which
does not define a loop in Γ or in Γ̂. Then composition with the edge collapse
Γ → Γ/e induces embeddings lj /e : S 1 → Γ/e such that (Γ/e, l/e) is again a
(n, k)-graph. By an edge collapse in an (n, k)-graph, we will always mean the
Algebraic & Geometric Topology, Volume 4 (2004)
550
Craig A. Jensen and Nathalie Wahl
collapse of an edge satisfying the above hypothesis. For a marked (n, k)-graph,
the marking of the collapsed graph is also obtained by composition with the
collapse.
A forest collapse in a (n, k)-graph (Γ, l) is a sequence of edge collapses. As the
order of the collapses does not matter, it is defined by the union of the edges
to be collapsed, which is a forest F in Γ. We denote the collapsed graph by
(Γ/F, l/F ).
Let Ln,k denote the set of equivalence classes of marked (n, k)-graphs. Define a
poset structure on Ln,k by saying that (Γ1 , φ1 ) ≤ (Γ2 , φ2 ) if there is a forest F
in Γ2 such that (Γ2 /F, φ2 /F ) is equivalent to (Γ1 , φ1 ). By abuse of notation, we
will also denote by Ln,k the simplicial complex which is the geometric realization
of this poset.
Σ
Using the geometric description of AΣ
n,k , we obtain a natural action of An,k
(and thus of An,k ) on Ln,k : For λ ∈ AΣ
n,k and (Γ, φ) ∈ Ln,k , define
λ · (Γ, φ) := (Γ, φ ◦ λ−1 ).
We consider two equivalence relations on (n, k)-graphs: Two (n, k)-graphs
(Γ1 , l 1 ) and (Γ2 , l 2 ) are Σ-equivalent if there is a graph isomorphism φ : Γ1 →
2 (◦) and φ◦l1 ' l2
Γ2 and a permutation σ ∈ Σk such that φ◦lj1 (◦) = lσ(j)
j
σ(j) (rel.
to ◦) for each j = 1, . . . , k . The (n, k)-graphs are equivalent if the permutation
σ is trivial.
Let QLΣ
n,k and QLn,k denote the set of all equivalence classes of (n, k)-graphs
under each equivalence relation. Define spaces QLΣ
n,k and QLn,k by taking
the elements of the sets as 0-simplices and attaching a p-simplex for each
equivalence class of chain of forests ∅ =
6 F1 ( F2 ( · · · ( Fp , where two chains
1
2
Fi and Fi are equivalent if there is a (Σ-)equivalence φ : (Γ, l) → (Γ, l) such
that φ(Fi1 ) = Fi2 for all i = 1, . . . , p. Note that these spaces are not simplicial
complexes in general. The space QLn,k is the moduli space of labeled (n, k)graphs, that is (n, k)-graphs with cycles labeled 1, . . . , k , whereas in QLΣ
n,k the
cycles are unlabeled.
Proposition 3.3 The action of AΣ
n,k (resp. An,k ) on Ln,k is transitive on the
sets of marked (n, k)-graphs with the same underlying unlabeled (n, k)-graph
(resp. same underlying labeled (n, k)-graph), with finite simplex stabilizers, and
the quotient of the action is QLΣ
n,k (resp. QLn,k ).
We will prove the following theorem in the next two sections.
Theorem 3.4 The space Ln,k is contractible.
Algebraic & Geometric Topology, Volume 4 (2004)
551
Automorphisms of free groups with boundaries
We use this result to give a range for the virtual cohomological dimension of
An,k . We recall that it is already known that the vcd of Aut(Fn ) = An,0 is
2n − 2 [5] and that the vcd of Ak0 is k − 1 [4]. Since An,k is a subgroup of
Aut(Fn+3k ) (see Proposition 5.1) and the latter group is virtually torsion free
and of finite vcd, the former group is as well [2, Example 1 on page 229].
Theorem 3.5
(1) 2n + 2k − 1 ≤ vcd(An,k ) ≤ 2n + 3k − 2
if k ≥ 1.
(2) vcd(An,1) = 2n + 1.
(3) vcd(A0,k ) = 2k − 1.
Proof of Theorem 3.5 For (1), we first note that there is a copy of Z2n+2k−1
in An,k : take n generators sending xi to xi yk and fixing all other generators,
then n generators sending xi to yk xi , then k conjugating yj by yk for 1 ≤ j ≤ k
and finally k − 1 conjugating yj by yj for j 6= k .
For an upper bound, we calculate the dimension of a maximal simplex in QLn,k .
Suppose the cycles lie in the maximally blown up graph in c connected components and that the graph has v vertices. The k + 1 vertices ∗, ◦1 , . . . , ◦k
all have valence 2. Since two cycles must meet in a valence 4 vertex and the
dual graph of the cycles in the graph is a forest, there are k − c vertices of
valence 4 in the graph. The remaining vertices are valence 3. Hence there are
e = 3(v − 2k + c − 1)/2 + (k + 1) + 2(k − c) = (3v − c − 1)/2 edges. Since
v − e = 1 − (n + k), this yields v = 2n + 2k + c − 1 vertices. We can collapse
these v vertices down to 1 by first collapsing all except one edge in each cycle
and then collapsing some remaining maximal tree. This yields a simplex of
dimension 2n + 2k + c − 2. Since c ≤ k , the dimension of a maximal simplex
is less than or equal to 2n + 3k − 2. The result follows from Theorem 3.4.
For (2), let k = 1 in (1).
Finally for (3) if n = 0 then c = 1 and so v = 2n + 2k + c − 1 = 2k . Hence a
maximal simplex has dimension 2k − 1.
In Figure 3, we show an example of a maximal simplex of QL1,2 given by a
graph with 2 connected cycle components and a forest consisting of 6 > 5 edges.
4
Initial segments of fixed point subcomplexes
Recall from [10] that the spine of Auter space Xn is a poset of pointed marked
graphs of genus n, where the marking is given by a map from the rose Rn :=
Algebraic & Geometric Topology, Volume 4 (2004)
552
Craig A. Jensen and Nathalie Wahl
◦1
∗
◦2
Figure 3: A maximal simplex in QL1,2 with a 6 edge forest
∨n S 1 . Xn is contractible and admits an action of Aut(Fn ). Let G be a finite
subgroup of Aut(Fn ) and consider the fixed point subcomplex XnG . A point in
XnG is a marked graph with a G-action. XnG is contractible [13, Thm. 1.1].
Edge collapses induce a poset structure on XnG with minimal elements the
reduced marked graphs, i.e. those with no G-invariant subforest. An edge
e of a marked G-graph Γ is essential if there exists a maximal G-invariant
forest which does not contain e. The graph Γ is essential if all of its edges are
essential. Let EnG be the full subcomplex of XnG spanned by essential marked
graphs. Collapsing inessential edges induces a poset map f : XnG → EnG such
that f (x) ≤ x for all x in XnG , and hence a deformation retraction on the
corresponding spaces by the Poset Lemma (see [27] or [4, Lem. 4.1] for a direct
proof of this special case).
Let W be the set of conjugacy classes of elements of Fn . Recall from [13,
Section 2] that, given any well ordering of W and of Fn , we can define a norm
k · ktot = k · kout × k · kaut ∈ ZW × ZFn
on EnG which well orders the reduced marked graphs of EnG (using the corresponding lexicographic ordering). For w ∈ W , the wth component of the
norm is the sum of the cyclic lengths of the paths φ(xw) for all x ∈ G, where
the cyclic length is the number of edges in the path reduced cyclically, i.e. the
length of the shortest path in its unpointed homotopy class. And for z ∈ Fn ,
the z th component of the norm is the sum of the lengths of the paths φ(xz) for
all x ∈ G, where we now consider the shortest paths in the pointed homotopy
classes.
A nonempty set Λ of reduced marked graphs of EnG is called an initial segment
if whenever two reduced marked graphs r1 , r2 satisfy kr1 ktot ≤ kr2 ktot and
r2 ∈ Λ, then r1 ∈ Λ. The realization of an initial segment Λ is the space
[
kΛk =
st(r),
r∈Λ
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
553
the union of the ascending stars (in EnG ) of the reduced marked graphs in Λ.
Let Λ be an initial segment and let Γ be a marked graph in kΛk. An edge e
of Γ is essential in Λ if there is a maximal G-invariant subforest F of Γ such
that e 6∈ F and Γ/F ∈ Λ. The graph Γ is essential in Λ if all of its edges
are essential in Λ. The essential realization of an initial segment Λ is the full
subcomplex |Λ| of kΛk spanned by the essential marked graphs of Λ.
Let Γ be a marked graph in EnG and let v be a vertex of Γ. An ideal edge γ at
the vertex v is a set of half edges of Γ terminating at v such that the blow-up
Γγ is again in EnG , where Γγ is the graph obtained by G-equivariantly pulling
the half edges in γ away from v (see [17, Section 5.A.] and [13, Section 2] for a
characterization of ideal edges when Γ is reduced.) More precisely, one creates
a new edge orbit Gγ and a new vertex orbit Gv(γ). In Γγ , γ is an actual edge
that goes from v(γ) to v . In addition, each half edge e ∈ γ is now attached to
v(γ) instead of v , and this process is done equivariantly on Gγ . Note that the
original graph Γ can be recovered from its blow up Γγ by collapsing Gγ .
An ideal forest in a reduced marked graph is a sequence of ideal edges satisfying
certain technical conditions, which make sure that by blowing up these ideal
edges, one can construct a blow-up ΓF in the star of Γ with an actual Gequivariant forest, whose partial collapses contain all the blow-ups of the ideal
edges in the ideal forest. One can define a poset structure on the set of ideal
forests of a reduced marked graph such that blowing up induces an isomorphism
between this poset and the star of Γ in EnG [17, Prop. 5.9].
Let γ be an ideal edge of a reduced G-graph Γ. Define D(γ) to be the set of
edges c of γ such that Gc is a forest in Γγ . To a pair (γ, c) with c ∈ D(γ)
corresponds a Whitehead move that consists of first blowing up γ and then
collapsing Gc. An ideal edge γ is reductive if there is a c ∈ D(γ) such that
performing the Whitehead move (γ, c) yields a graph of smaller norm.
Let Y be any of EnG , kΛk or |Λ|. The reductive link of a reduced marked graph
Γ in Y is the part of the star of Γ in Y spanned by ideal forests with at least
one reductive ideal edge. The purely reductive link of Γ in Y is the part of the
star of Γ in Y spanned by nontrivial ideal forests with all edges reductive. By
the Poset Lemma, there is a deformation retraction from the reductive link of Γ
to the purely reductive link of Γ given by collapsing non-reductive ideal edges.
Theorem 4.1 If Λ is an initial segment of EnG , then both its realization kΛk
and its essential realization |Λ| are contractible.
Algebraic & Geometric Topology, Volume 4 (2004)
554
Craig A. Jensen and Nathalie Wahl
Proof By the Poset Lemma, collapsing inessential edges induces a deformation
retraction from kΛk to |Λ|. Thus it is enough to show that kΛk is contractible.
If r is a reduced marked graph in kΛk, we have already observed that the
reductive link of r in kΛk deformation retracts onto the purely reductive link
of r in kΛk. Since Λ is an initial segment, the purely reductive link of r in kΛk
is the purely reductive link of r in EnG . From [13, Lem. 3.1 - 3.7], this reductive
link (referred to as S(R) in [13]) is contractible. As Λ is an initial segment, all
Γ0 such that kr 0 k < krk are also in Λ. The reductive link of a reduced marked
graph r is the intersection
\
red(r) = stkΛk (r) ∩
stkΛk (r 0 ).
kr 0 k<krk
Now the norm well orders the reduced marked graphs in kΛk. We can build
kΛk as the union of the stars of the elements of Λ, starting with the star of
the smallest element and adding the stars successively. As the star of a graph
is contractible and all the successive intersections are contractible, transfinite
induction implies that every connected component of kΛk is contractible.
It remains to show that kΛk is connected. From [16, Thm. 2], we know that any
two reduced marked graphs in EnG are connected by a sequence of Whitehead
moves. The reduced marked graphs of EnG are ordered by the norm. Choose
I ⊆ ZW × ZFn such that the ordered set of reduced marked graphs of EnG is
{ri : i ∈ I} and denote by r0 the least element. Note that r0 ∈ Λ because Λ
is nonempty. Note also that the stars stkΛk (ri ) and stEnG (ri ) of ri in kΛk and
EnG are equal. We will denote these stars by st(ri ).
Suppose that kΛk is not connected. The star st(r0 ) is contractible and thus
connected. Take the least index j such that ∪i≤j st(ri ) is not connected. Then
we can get from r0 to any ri with i < j in kΛk, but not from r0 to rj . Since
EnG is connected and I is a well ordering, we can choose a least k such that
there is a path from r0 to rj in ∪i≤k st(ri ). This implies that the reductive
link of rk is not connected, however, since it has at least two components: one
coming from r0 and the other coming from rj . This contradicts the fact that
the reductive link is contractible. So kΛk is connected.
5
Contractibility of Ln,k
In this section, we prove Theorem 3.4 by identifying Ln,k with the essential
realization of an initial segment and applying Theorem 4.1.
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
555
Let m = n + 3k . Let {x1 , . . . , xn , y1 , . . . , yk , u1 , . . . , uk , v1 , . . . , vk } denote the
generators of Fm . For each j = 1, . . . , k , let Θj ∼
= Z/3 be the subgroup of
Aut(Fm ) generated by θj , where θj (uj ) = vj , θj (vj ) = vj−1 u−1
j , and all other
generators are fixed. This group can be realized as the group of rotations of
the three edges of a theta-graph (see Figure 5). Let
Θ = Θ1 × Θ2 × · · · × Θk .
be the direct product of these k groups. The normalizer NAut(Fm ) (Θ) acts on
Θ and thus on E Θ . Both spaces are contractible
the fixed point subcomplex Xm
m
[13] (see also Section 4).
Note that a similar fixed point subcomplex was used in [15] to study the holomorph of free groups Fn o Aut(Fn ). This last group is the subgroup of A1n of
automorphisms where y1 does not occur in the images of the xi ’s. One could
work with circles and a Z/2-action, instead of Θ-graphs and a Z/3-action.
This would however bring a few extra technical problems (having to introduce
extra vertices to avoid acting on edges by inversions) while producing the same
complex. (See also Section 6.)
Although An,k is not a subgroup of Aut(Fn+k ), it is a subgroup of Aut(Fn+3k ):
Proposition 5.1 There is an inclusion of groups β : AΣ
n,k ,→ NAut(Fn+3k ) (Θ) 6
Aut(Fn+3k ).
Proof The map β is induced geometrically by attaching two additional circles
at the end of each of the k stems (see Figure 4). Algebraically, this map can
be described as follows: for hσ, νi , wj i ∈ AΣ
n,k = Σk o An,k ,

xi → ν i


 y
→ wj−1 yσ(j) wj
j
φ = ihσ, νi , wj i maps
uj → wj−1 uσ(j) wj



vj → wj−1 vσ(j) wj
One then checks easily that φ = βhσ, νi , wj i satisfies φ θl φ−1 = θσ(l) for each
l = 1, . . . , k .
For each j = 1, . . . , k , let θ j denote a graph with two vertices ◦j and vj ,
and three edges ej1 , ej2 , ej3 where each eji goes from ◦j to vj (See Figure 5).
Consider the subcomplex F ix of Xm spanned by marked graphs φ : Rn → Γ̃
which satisfy the following conditions:
Algebraic & Geometric Topology, Volume 4 (2004)
556
Craig A. Jensen and Nathalie Wahl
u1
v1 yk
uk
y1
vk
x1
xn
x2
Figure 4: Graph giving the inclusion An,k ,→ Aut(Fn+3k )
• The graph Γ̃ is obtained from a graph Γ of genus n + k by attaching the
graphs θ 1 , . . . , θ k to Γ, identifying the vertex ◦j of θ j with a vertex of Γ
and remembering these —possibly multiple— labels.
• For each j = 1, . . . , k , there exists a path wj in Γ̃ from ◦j to ∗ such that
φ(uj ) = w̄j ? ej1 ? ēj2 ? wj
and
φ(vj ) = w̄j ? ej2 ? ēj3 ? wj
where p̄ denote the path p with the reverse orientation and ? denotes
the concatenation of paths.
vj
ej1
ej2
ej3
◦2
◦1 = ◦3
◦j
Figure 5: The graph θj and a graph of F ix
Θ ∼ F ix ∼ X Θ .
Proposition 5.2 Em
=
= m
Proof Each marked graph in F ix is fixed by the action of Θ as θj cyclically
Θ.
rotates the edges of the attached θ j . So F ix ⊂ Xm
Θ be the graph obtained by wedging the k graphs θ j to the rose R
Let Rn,k
n+k .
Recall that a graph is reduced if it has no invariant subforest. Note that all the
Θ .
reduced graphs in F ix have the same underlying graph Rn,k
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
557
Θ are connected by a seKrstić showed that any two reduced Θ-graphs in Xm
quence of Whitehead moves ({e, f }, f ) where e, f have the same terminal vertex
and StabΘ (e) ⊆ StabΘ (f ) (and e ∈
/ Θf ) [16, Prop. 4’]. Pictorially the move
({e, f }, f ) corresponds to pulling each edge in Θe along the corresponding edge
in Θf . Because of the condition on stabilizers, we can only perform Whitehead
Θ of the form ({e, f }, f ) or ({ēj , f }, f ) where e, f are edges of
moves on Rn,k
i
Rn+k and ēji is in θ j . The first type of moves affect only the first n + k petals,
whereas the second type takes the subgraph θ j around the edge f . The reduced
graphs in F ix are closed under these moves.
Θ are exactly the reduced graphs in F ix. Now
Hence the reduced graphs in Xm
the ascending star of a reduced graph in F ix is the same as the ascending star of
Θ as equivariant blow-ups preserve the θ -graphs (see [12, Claim
the graph in Xm
Θ . Note finally that all graphs
6.6] for a similar computation). So F ix = Xm
in F ix are essential as no edge in a θ j can be collapsed Θ-equivariantly and
the other edges, being non-separating by hypothesis, are thus also essential. So
Θ = EΘ .
F ix = Xm
m
We now choose a well ordering of ZW × ZFn such that the the space Ln,k of
marked (n, k)-graphs is the realization of an initial segment of the induced norm
k · ktot . Well order the set of conjugacy classes W so that its first k elements
are
u 1 y 1 u1 , u 2 y 2 u2 , . . . , u k y k uk
and well order Fn in any manner. Now define Λ to be the set of reduced ΘΘ whose norm is minimal in the first k coordinates. This is an
graphs in Em
initial segment in the lexicographic ordering induced by the norm. Recall that
Θ have underlying graph RΘ . The minimal
all reduced marked graphs in Em
n,k
norm on the component uj yj uj is thus (2 + 1 + 2) · |Θ| = 5 · 3k . The marking
φ of such a minimal reduced marked graph is determined on yj , uj and vj by
a loop wj at ∗ and a single edge loop Sj at ∗ in the graph as follows:
φ(yj )
φ(uj )
φ(vj )
=
=
=
w̄j ? Sj ? wj
w̄j ? ej1 ? ēj2 ? wj
w̄j ? ej2 ? ēj3 ? wj ,
Note that the edge loops Sj have to be different for φ to be a homotopy equivalence.
By Theorem 4.1, both kΛk and |Λ| are contractible. So Theorem 3.4 follows
directly from the following statement:
Theorem 5.3 There is a poset equivalence between Ln,k and |Λ|.
Algebraic & Geometric Topology, Volume 4 (2004)
558
Craig A. Jensen and Nathalie Wahl
Lemma 5.4 Let e be an edge in a (n, k)-graph (Γ, l). Then there exists
a maximal forest in Γ − e, that is a forest F such that Γ/F is a (reduced)
(n, k)-graph.
Proof Let Γ̂ be the graph of genus n obtained from Γ by collapsing the cycles.
Take a maximal forest F̂ in Γ̂. It can be extended to a maximal forest F of Γ
by picking all but one edge in each cycle.
If e is an edge in a cycle Sj , we can choose such a maximal forest containing
Sj − {e} to obtain the result. If e is not in a cycle, e is non-separating in Γ by
assumption and hence non-separating in Γ̂, so we can choose a forest F̂ in Γ̂
which does not contain e and extend it to Γ as above.
Proof of the Theorem Let (Γ, φ) be a marked (n, k)-graph. Its underlying
(n, k)-graph has Sj = φ(Cj ) with basepoint ◦j , and let wj be the path from ◦j
to ∗ in Γ, image of the stem of the j th cycle in Gn,k , so that φ(yj ) = w̄j ?Sj ?wj .
Define a graph Γ̃ by attaching θ j to ◦j in Γ for each j and define a marking
φ̃ : Rn+3k → Γ̃ by
φ̃(xi ) = φ(xi )
φ̃(yj ) = w̄j ? Sj ? wj
φ̃(uj ) = w̄j ? ej1 ? ēj2 ? wj
φ̃(vj ) = w̄j ? ej2 ? ēj3 ? wj .
This defines an inclusion of Ln,k into kΛk. Indeed, the poset minimal elements
of Ln,k are mapped to Λ, so their stars are in kΛk.
To show that it maps into the smaller subcomplex |Λ|, we need to show that
for each (Γ, φ) in Ln,k , the pair (Γ̃, φ̃) is essential in Λ, that is for each edge
e of Γ̃, there is an invariant forest F ∈ Γ̃ − e such that Γ̃/F ∈ Λ. Note first
that a maximal invariant forest of Γ̃ has to lie in Γ. If e lies in θ j for some j ,
choose a maximal forest F in Γ such that Γ/F is a reduced (n, k)-graph and
hence Γ̃/F is in Λ. On the other hand, if e lies in Γ, we know that such a
forest exists in Γ/e by Lemma 5.4.
We claim that the image of Ln,k equals |Λ|. To show this, we need to show
that the marking of any graph in |Λ| is induced by a marking from Gn,k to
the graph minus the theta subgraphs, that is we need to show how that the
image of yj can be decomposed as a cycle with a stem for each j with at most
a point in the intersection of any two different cycles, and we need to show that
the uj ’s and vj ’s are mapped accordingly. This is true for the poset minimal
elements, i.e. the elements of Λ: the cycles are the Sj ’s with basepoint ◦j = ∗
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
559
and with stem wj . So we need to show that the cycle structure is carried by
the admissible blow-ups (that is the blow-ups which stay within |Λ|).
Let (Γ, ψ) ∈ |Λ| be a graph which has a cycle structure and let γ be an ideal
edge at a vertex v of Γ such that Γγ is still in |Λ|. We want to show that Γγ
has a cycle structure.
Case 1 Suppose that v 6= ◦j for all j , so there is no θ j attached to v . Each
cycle has at most two edges ending at v . If γ has either 0 or 2 half edges from
Sj for each j except for at most one j , then the induced marking on Γγ still
has embedded circles, with intersection possibly smaller than in Γ.
Suppose now that there is j 6= j 0 such that both j and j 0 have exactly one half
edge in γ . Then the induced cycles in Γγ are such that Sj ∩ Sj 0 = γ , where
γ is now the new edge in Γγ . But there is no forest F in Γγ − {γ} such that
Γγ /F is in Λ. Indeed, if we do not collapse γ again, the induced marking will
have the j th and j 0 th cycles intersecting in γ . Hence Γγ is not in |Λ|.
Case 2 Suppose that there is at least one j such that v = ◦j . In particular,
there is at least one θ -graph attached to v . Note first that the equivariance
conditions on γ imply that at most one half edge of a graph θ j may be in γ ,
and if there is such an edge, the whole θ -graph is moved along the new edge γ
in Γγ . Moreover, by the same argument as in Case 1, their can be at most one
j such that Sj has exactly one half edge in γ .
Suppose that there is a j such that Sj has two half edges in γ . Then eji must
also be in γ for some i, otherwise any forest collapse which does not collapse γ
will all induce a marking with w̄j ? γ ? Sj ? γ̄ ? wj for yj but still w̄j ? ej1 ? ēj2 ? wj
for uj . Similarly, if a half edge eji is in γ for some j , then the two half edges
of Sj terminating at v must also be in γ . This shows that a blow up cannot
disconnect θj from Sj .
Lastly, if there is a j such that only one half edge is in γ , then the induced
marking will move ◦j and prolong wj if there is also an eji in Γ.
6
A space Lkn for the group Akn
We recall from Section 2 that there is a central extension
Zk → An,k → Akn ,
where Akn is a subgroup of Aut(Fn+k ). To this group extension, corresponds a
sequence of contractible spaces
Fn,k → Ln,k → Lkn
Algebraic & Geometric Topology, Volume 4 (2004)
560
Craig A. Jensen and Nathalie Wahl
where the spaces are realizations of posets, the maps are poset maps and Fn,k ∼
=
Rk . We describe here the spaces Lkn and Fn,k .
As Akn is a subgroup of Aut(Fn+k ), a natural way to construct a contractible
space Lkn with an action of Akn with finite stabilizers and finite quotient, is to
consider a subcomplex of Auter space Xn+k , as small as possible to get as good
an approximation as we can of the vcd. In the same spirit as the construction
of Ln,k , one can consider the essential realization |Λ0 | of the initial segment Λ0
of reduced graphs of minimal norm with respect to the n-tuple of cyclic words
(y1 , . . . , yk ). We claim that |Λ0 | is homeomorphic to the following space Lkn :
Definition 6.1 An nk -graph (Γ, l) is a graph Γ of genus (n + k) with vertices
of valence at least 3, except the basepoint which has valence at least 2, with no
separating edges, and equipped with embeddings lj : S 1 → Γ for j = 1, . . . , k
such that any two Sj := lj (S 1 ) meet in at most a point and the dual graph of
the Sj ’s is a forest.
A marked nk -graph (Γ, φ) is a graph Γ equipped with
a pointed homotopy
n
equivalence φ : Gn,k → Γ such that (Γ, φ ◦ κ) is an k -graph.
Two marked nk -graphs (Γ1 , φ1 ) and (Γ2 , φ2 ) are equivalent if there is a graph
isomorphism ψ : Γ1 → Γ2 such that ψ ◦ φ1 (Cj ) = φ2 (Cj ) and ψ ◦ φ1 ' φ2 (rel.
Cj for all j , up to a rotation).
Define now Lkn to be the poset of equivalence classes of marked nk -graphs (or
its realization) where the poset structure is given by the same admissible edge
collapses as for Ln,k (see p6).
Theorem 6.2 There is a poset equivalence Lkn → |Λ0 | and hence Lkn is contractible.
The proof is totally analogous to the proof of Theorem 5.3, although a little
bit simpler as we do not need to carry around Θ-graphs. The poset map
Lkn → Xn+k takes the minimal poset elements of Lkn to Λ0 , and Lkn maps into
the essential realization of Λ0 by Lemma 5.4. On the other hand, we have
Λ0 ⊂ Lkn and the inclusion of |Λ0 | reduced to the case 1 in the proof of Theorem
5.3. Finally, |Λ0 | is contractible by Theorem 4.1 with G the trivial group.
Proposition 6.3 The quotient space Lkn /Akn := QLkn has dimension 2n+2k−2
if n ≥ 1 and dimension k − 1 if n = 0.
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
561
Again, the proof is essentially the same as for QLn,k (in Theorem 3.5).
Note that Collins constructed a space for Ak0 which is isomorphic to our space
Lk0 (up to basepoint questions) [4]. In this particular case when n = 0, the
dimension is equal to the vcd.
There is a forgetful map Ln,k → Lkn . Let Fn,k be the fiber over the rose
(Rn+k , id) with the Zk -action induced by the action of An,k on Ln,k .
Proposition 6.4 There is a poset equivalence Fn,k → Zk × (Z/2)k , where the
poset structure on Zk × (Z/2)k is defined by (z̄, p̄) ≤ (z̄ 0 , p̄0 ) iff for all i we have
pi = p0i and zi = zi0 or
pi < p0i and zi = zi0 or zi = zi0 + 1.
Moreover, the realization of the poset Zk × (Z/2)k is homeomorphic to Rk and
the Zk -action on Fn,k translates to the canonical Zk -action on Rk .
Proof Suppose (Γ, φ) ∈ Fn,k . Then Γ is the rose Rn+k with additional basepoints for the cycles, which are either equal to the basepoint of the graph, or
a valence two vertex on the petal. Up to (n, k)-graph isomorphisms, there are
(Z/2)k possibilities. We identify p̄ ∈ (Z/2)k with the rose which has a valence
two vertex on the (n + j)th petal if pj = 1 and no such vertex if pj = 0.
Now φ : Gn,k → Rn+k is homotopic to the identity if the cycles are allowed
to rotate. So there are Zk possibilities for φ, where z̄ ∈ Zk for a rose (R, p̄)
represents the map φ which maps the cycles via the identity, according to the
basepoints parametrized by p̄, and wraps the j th stem zj times around the
j th cycle.
If pj = 0 for all j , then the graph is reduced. If there exists a j such that
pj = 1, then there are two possible collapses on the j th petal, one which leaves
zj constant, and one which increases zj by 1. (We choose the identification
φ
z̄ so that this is the case.) This corresponds to the poset structure of
Zk × (Z/2)k .
We are left to show that the realization of the poset Zk × (Z/2)k is homeomorphic to Rk and that the induced Zk action is the canonical action. Embed the
set Zk ×(Z/2)k in Rk by mapping (z̄, p̄) to (z̄ + 2p̄ ) = (z1 + p21 , . . . , zk + p2k ) ∈ Rk .
Let Cz̄ be the cube of length 1 in Rk with bottom corner z̄ and top corner
z̄ + 1̄. Let cz̄ = (z1 + 12 , . . . , zk + 12 ) be the center of the cube. Then we have
Cz̄ ∩Fn,k = {(z̄, p̄) ∈ Fn,k |(z̄, p̄) ≤ cz̄ = (z̄, 1̄)}. Similarly, the center of each face
is the maximal element of the face in the poset. It follows that the realization
of Zk × (Z/2)k is Rk . The Zk -action, which is by addition on the z̄ -coordinate,
is then by translation.
Algebraic & Geometric Topology, Volume 4 (2004)
562
Craig A. Jensen and Nathalie Wahl
The contractibility of Lkn and Fn,k leads to an alternative proof of the contractibility of Ln,k . This second approach, suggested to us by the referee,
makes more apparent the fact that the Θ-graphs in the original proof of the
contractibility of Ln,k only play the role of adding basepoints.
7
Presentation
We denote by Pi,j the automorphism of the free group Fn = hx1 , . . . , xn i which
permutes xi and xj and leaves the other generators fixed, and we denote by Ii
the automorphism that maps xi to its inverse and fixes the other generators.
For Whitehead moves, we use the following shortened notation: for , η = ±1,
• (zi ; zjη ) denotes the automorphism which maps zi to zi zjη if = 1 and to
zj−η zi if = −1, and fixes the other generators;
• (zi± ; zjη )denotes the automorphism which maps zi to zj−η zi zjη and fixes the
other generators.
In terms of the discussion of Whitehead moves at the beginning of Section 4
before Theorem 4.1, (zi ; zjη ) := ({zi , zjη }; zjη ) and (zi± ; zjη ) := ({zi , zi−1 , zjη }; zjη ).
Note that (zi ; zjη )−1 = (zi ; zj−η ) and (zi± ; zjη )−1 = (zi± ; zj−η ). We will use the
second notation when writing relations for simplicity.
Theorem 7.1 The following gives a presentation for Akn :
Generators:
Pi,j
for
Ii
for
(xi ; z) for
(yi± ; z) for
1 ≤ i, j ≤ n and i 6= j
1≤i≤n
1 ≤ i ≤ n, = ±1 and xi 6= z ∈ {x1 , . . . , xn , y1 , . . . , yk }
1 ≤ i ≤ k and yi 6= z ∈ {x1 , . . . , xn , y1 , . . . , yk }
Relations: For z, zi ∈ {x1 , . . . , xn , y1 , . . . , yk } and w, wi = xδjii or yj±i ,
Q1
Q2
Q3
Q4
Q5
Relations in Aut(Fn ) for {(xi ; xj ), Pi,j , Ij }
(w1 ; z1 )(w2 ; z2 ) = (w2 ; z2 )(w1 ; z1 ) for w1 6= w2 and zi±1 ∈
/ {w1 , w2 }
(1) (yi± ; xj ) Pj,l = Pj,l (yi± ; xl )
(2) (yi± ; xj ) Ij = Ij (yi± ; x−1
j )
η
η
−η
(1) (w; xj )(xj ; z)(w; xj ) = (w; z)(xηj ; z)
(2) (yi± ; z )(w; yi )(yi± ; z − ) = (w; z − )(w; yi )(w; z )
η −1
± η
(yi± ; xηj )(x−η
j ; yi ) = (xj ; yi )(yi ; xj )
whenever these symbols denote generators or their inverses.
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
563
In particular, for n = 0 we recover McCool’s presentation of the pure symmetric
(or basis-conjugating) automorphisms of free groups [25], which has generators
{(yi± ; yj ) | 1 ≤ i, j ≤ k, i 6= j} and relations
Z1 (yi± ; yj )(yl± ; yj ) = (yl± ; yj )(yi± ; yj )
Z2 (yi± ; yj )(yl± ; ym ) = (yl± ; ym )(yi± ; yj )
Z3 (yi± ; yj )(yl± ; yj )(yi± ; yl ) = (yi± ; yl )(yi± ; yj )(yl± ; yj )
where the indices are assumed to be different when given by different letters.
Theorem 7.2 The following gives a presentation for An,k :
Generators: same as for Akn with k additional generators (yi± ; yi ) for 1 ≤ i ≤ k .
Relations: Q1, Q2, Q3, Q4 (never allowing a symbol (yi± ; yi )), and
Q2 0
Q5 0
(yi± ; yi )a = a(yi± ; yi ) with a = Pi,j , Ij or (w; z)
η
± −η
± −1
(yi± ; xηj )(x−η
j ; yi )(yi ; xj )(xj ; yi ) = (yi ; yi )
whenever these symbols denote generators or their inverses.
The generator (yi± ; yi ) := (x1 , . . . , xn , 1, . . . , yi , . . . , 1) ∈ An,k should be thought
of as the conjugation of yi by itself, even though this is trivial as an automorphism of the free group.
Proof of Theorem 7.1 The group Aut(Fn+k ) acts on the set of cyclic words
−1
k
with letters in Ln+k := {x1 , x−1
1 , . . . , yk , yk } and An is the stabilizer of the
k -tuple of cyclic words U = (y1 , . . . , yk ). Theorem 1.1 in [24] says that such a
group is finitely presented. As remarked in [24, Section 4(1)], the proof actually
constructs a finite 2-dimensional complex K whose fundamental group is this
stabilizer and hence gives a presentation. We analyze this complex.
The vertices of the complex in our case are
K0 := {V minimal tuple |V = U P for some P ∈ Aut(Fn+k )} = {U σ|σ ∈ Ωk }
where a minimal tuple is a tuple of reduced cyclic words of minimal length in
the orbit of U and Ωk ∼
= Σk o Z2 is the subgroup of Aut(Fn+k ) of permutations
and inverses of the yi ’s. So K0 ∼
= Ωk .
Let Wn+k denote the set of Whitehead automorphisms for Aut(Fn+k ). We
recall that Wn+k is the finite set T1 ∪ T2 where T1 = Ωn+k is the set of
permutations and inverses and T2 = {(A; a)|A ⊂ Ln+k , a ∈ A and a−1 ∈
/ A}
−1
with (A; a) denoting the automorphism mapping z ∈ Ln+k to za (and z to
Algebraic & Geometric Topology, Volume 4 (2004)
564
Craig A. Jensen and Nathalie Wahl
a−1 z −1 ) if a 6= z ∈ A and z −1 ∈
/ A, mapping z to a−1 za (and z −1 to a−1 z −1 a)
−1
if z, z ∈ A and fixing z otherwise. The set of edges in the complex is
K1 := {(V1 , V2 ; P )|P ∈ Wn+k and V1 P = V2 }/ ∼ .
Note that if P ∈ W , then P −1 is also in W . The directed edge e = (V1 , V2 ; P ),
from V1 to V2 , is identified with ē = (V2 , V1 ; P −1 ) in K1 . Finally the 2-cells are
all the cells coming from the relations R1 to R10 in [23], attached to directed
paths. We recall these relations below.
There are two types of edges in K , coming from the two types of Whitehead
automorphisms:
Type I edges (U σ, U στ ; γ) for σ, τ ∈ Ωk and γ ∈ Ωn+k such that (U σ)γ =
U στ . In particular, γ is only allowed to permute the yi ’s among themselves.
Hence γ ∈ Ωn × Ωk ⊂ Ωn+k and γ = (γ, τ ). So the set of type I edges between
any two vertices is isomorphic to Ωn .
Type II edges Type II edges do not involve permutations so they can exist
only from one vertex to itself in our complex. Moreover, they have to preserve
the cyclic words y1 , . . . , yk , so they have the form
(U σ, U σ; (X t {yi1 , yi−1
, . . . , yir , yi−1
} t {a}; a)
r
1
−1
where X ⊂ Lx = {x1 , x−1
1 , . . . , xn , xn } and a ∈ Ln+k .
Choose a maximal tree T in K1 with edges of type (U σ, U στ ; τ ) for some
τ ∈ Ωk ∼
= {1} × Ωk . The generators of π1 K ∼
= Akn are given by the edges in
K1 − T .
If we consider the subcomplex of K formed by taking all the edges of the form
(U σ, U στ ; τ ) for τ ∈ Ωk and the 2-cells given by the relations R7 (which is a
set of relations for Ωn+k ) restricted to these edges, we obtain a retraction of
the 2-skeleton of EΩk , which is simply connected. It follows that all edges of
type (U σ, U στ ; τ ) give trivial generators in Akn .
Moreover, (U σ, U στ ; (γ, τ )) ∼ (U, U ; (γ, 1)) as (1, σ)(γ, τ ) = (γ, 1)(1, στ ) in
Ωn+k (giving thus a relation in R7) and (U σ, U σ; (A; a)) ∼ (U, U ; (Aσ −1 ; aσ −1 ))
as (1, σ)(A; a) = (Aσ −1 ; aσ −1 )(1, σ) by R6. (R6 says that T −1 (A; a)T =
(AT ; aT ) for any T ∈ T1 .) So the loops at the vertex U generate the group.
Also, the relations given by 2-cells in the complex all follow from the relations
R1-R10 interpreted on the generators (U, U ; γ) and (U, U ; (A; a)). Indeed, the
other relations are conjugated to these (except for the relations in R6 and R7
already used to identify generators) and thus do not bring anything new by R6,
used in the case T = (1, σ).
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
565
R1 identifies the inverses in T2 and R2 says that (A; a)(B; a) = (A ∪ B; a)
when A ∩ B = {a}, showing thus that the generators we propose are indeed
generators for the group. Note that R2 also implies the commutation relations
of Q2 when z1 = z2 . As the relations in the Theorem are easily verified, we are
left to show that they imply the relations R3-R10.
R3 is a set of commutation relations and follows from Q2 when z1 6= z2 .
R4 says that (B; b)−1 (A; a)(B; b) = (A + B − b; a) when A ∩ B = ∅, a−1 ∈
/B
−1
and b ∈ A. We need to deduce it from our relations when (A; a) and (B; b)
define loops at U in K . In particular, we need b = xi ∈ Lx as b−1 ∈ A but
b∈
/ A and b 6= a−1 . Using Q2, we can isolate a minimal case on the left hand
−
−
side: (B − w; xi )−1 (w; x−
i )(xi ; a)(w; xi )(A − xi ; a)(B − w; xi ). We then use
Q4 (1) and repeat the procedure with all the elements of B .


a b
R5 says that (A; a)(A − a + a−1 ; b) =  ↓ ↓  (A − b + b−1 ; a) when b ∈ A,
b−1 a
b−1 ∈
/ A and a 6= b. In our case, we need to have a, b ∈ Lx . The relation follows
from Q1 when only x’s are involved. Using R2, we isolate (b; a)(a−1 ; b) in the
left hand side and use R5 in this minimal case (that is use Q1). We then use
Q3 to move the matrix (element of T1) in front of the left hand side and finally
use R4.
R6, which was recalled above, follows from Q1 and Q3 (using Q2 as usual).
R7 gives a set of relations for Ωn , which is included in Q1.
R8 says that (A; a) = (Ln+k −a−1 ; a)(Ln+k −A; a−1 ) = (Ln+k −A; a−1 )(Ln+k −
a−1 ; a). This just follows from R1-R2.
R9 says that (Ln+k − b; b−1 )(A; a)(Ln+k − b−1 ; b) = (A; a) if b, b−1 ∈
/ A. In
δ
particular, a 6= b. This relation follows from Q4. Indeed, if a = xj ∈ Lx , isolate
−1
δ
(x−δ
j ; b ) from of the left hand side and use Q4 (1) to pass it over (A; xj ), which
−δ −1
−δ
gives (Ln+k − b − xj ; b )(A; xδj )(A − xδj + b−1 ; b−1 )(xj ; b−1 )(Ln+k − b−1 ; b).
Now simplify and isolate (xδj ; b) and use again Q4(1) to pass it over (A; xδj ),
−δ
−1
δ
δ
δ
which gives (Ln+k − b − x−δ
j ; b )(xj ; b)(A − xj + b; b)(A; xj )(Ln+k − A − xj −
b−1 ; b).
This simplifies to (A; xδj ) as required.
If on the other hand a = yi ∈ Ly , isolating (yi± ; b−1 ) in the left hand side and
using Q4 (2) to pass it over (A; yi ) gives
Algebraic & Geometric Topology, Volume 4 (2004)
566
Craig A. Jensen and Nathalie Wahl
(Ln+k −b−yi± ; b−1 )(A−yi +b; b)(A; yi )(A−yi +b−1 ; b−1 )(yi± ; b−1 )(Ln+k −b−1 ; b),
which simplifies to (A; yi ).
R10 says that (Ln+k − b; b−1 )(A; a)(Ln+k − b−1 ; b) = (Ln+k − A; a−1 ) if b 6= a,
b ∈ A and b−1 ∈
/ A. So b ∈ Lx in our case. If a ∈ Lx , we can follow [23] and
show that the relation follows from R9 and R5. We do the case a ∈ Ly . So we
take a = yjη , b = xi The left hand side in the equation equals
± −
− η
η
(Ln+k − xi − yj± ; x−
i )(yj ; xi )(xi ; yj )(A − xi ; yj )(Ln+k − xi ; xi )
which, using Q5, gives
− −η
± −
− η
(Ln+k − xi − yj± ; x−
i )(xi ; yj )(yj ; xi )(A − xi ; yj )(Ln+k − xi ; xi ).
−η
± −
Now we use Q4 (1) to pass (x−
i ; yj ) to the left, and Q4(2) to pass (yj ; xi )
η
over (A − xi ; yj ), which gives
−η
η −η
η ± −
η
(x−
i ; yj )(Ln+k − xi − yj ; yj )(Ln+k − xi − yj ; xi )(A − yj ; xi )(A − xi ; yj )
η
−
± −
− (A − xi − yj + x−
i ; xi )(yj ; xi )(Ln+k − xi ; xi ).
This simplifies to (Ln+k − A; yj−η ) as required.
To prove Theorem 7.2, we need a lemma about group extensions.
i
π
Let 1 → A → E → G → 1 be a central extension of a group G by an abelian
group A, with a section of sets s : G → E (so π ◦ s = id) satisfying s(1) = 1
and f : G × G → A the classifying cocycle of the extension, defined by
s(g) s(h) = i◦f (g, h) s(gh).
Note that if r1 . . . rm = 1 in G, then s(r1 . . . rm ) = 1 in E and hence
s(r1 ) . . . s(rm ) = f (r1 , r2 )f (r1 r2 , r3 ) . . . f (r1 . . . rm−1 , rm ).
We denote by sf the operator lifting the relations of G to relations in E this
way.
i
π
Lemma 7.3 Let 1 → A → E → G → 1 be a central extension as above. If A
and G are finitely presented, with presentation
A = ha1 , . . . , ak i/RA
G = hg1 , . . . , gn i/RG .
Then E is finitely presented, with presentation
E = hi(a1 ) . . . , i(ak ), s(g1 ), . . . , s(gn )i/[A, G] ∪ i(RA ) ∪ sf (RG )
where [A, G] is the set of commutation relations i(aj )s(gi ) = s(gi )i(aj ) and
sf (RG ) is the lift of the relations RG to E twisted by f as defined above.
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
567
Proof Consider the free group F := hi(a1 ) . . . , i(ak ), s(g1 ), . . . , s(gn )i. The
natural map h : F → E is surjective as i × s : A × G → E is a bijection of sets
(since s(G) is a set of coset representatives). The relations [A, G], i(RA ) and
sf (RG ) are clearly satisfied in E , and so define elements in the kernel of h.
Let e = i(aj1 ) . . . i(ajp )s(gi1 ) . . . s(giq ) ∈ F and suppose that h(e) = 1. (It is
enough to consider this case as [A, G] is satisfied in E .) Then π ◦h(e) = 1 in G,
that is gi1 . . . giq = 1. This relation in G lifts to a relation s(gi1 ) . . . s(giq ) = i(a)
in E for some a ∈ A. So i(aj1 ) . . . i(ajp )s(gi1 ) . . . s(giq ) = i(aj1 ) . . . i(ajp )i(a).
As h(e) = 1, we have aj1 . . . ajp a = 1 is a relation in A (as i is an injective
homomorphism). So the relation h(e) = 1 follows from a relation in sf (RG )
and a relation in i(RA ).
Proof of Theorem 7.2 There is a central extension 1 → Zk → An,k →
Akn → 1. By the Lemma, the generators of An,k are the union of the generators of Akn with those of Zk . The last generators are denoted (yi± ; yi ), for
1 ≤ i ≤ k , in the Theorem. Then relations for An,k are given by the relations
for Zk , the commutations [Zk , Akn ] and the lifts of the relations from Akn . Q2’
takes care of the first two sets of relations and Q1,Q2,Q3,Q4 and Q5’ is the lift
of the relations from Akn .
In [3], the group Ak0 is studied (denoted Ck in the paper). They show that
M
H1 (Ak0 ) =
Z[yi± ; yj ]
1≤i,j≤k, i6=j
where Z[yi± ; yj ] is the free abelian group generated by [yi± ; yj ]. Recall that Ak0
can be thought of as an extended pure braid group (which is the mapping class
group of a punctured sphere). Note in particular that, as for the pure braid
group, H1 is not stable with respect to k . The presentation given above allows
us to calculate H1 in the higher genus cases.
Theorem 7.4 H1 (An,k ) ∼
= H1 (Akn ) if n > 2
= Z2 ∼
Proof If n > 2, we have H1 (Aut(Fn )) = Z2 generated by [Pi,j ] = [Ik ] (the
generators (xδi ; xj ) being trivial in the abelianization).
For Akn , Q4 (1) implies that (w; z) is trivial in the abelianization for any w = xδi
or yi± and z ∈ {x1 , x2 , . . . , yk }. The only relations involving Pi,j or Ik which are
not in Aut(Fn ) give trivial relations in the abelianization, hence H1 (Akn ) = Z2 .
For An,k , the additional relation Q5’ implies that the generators (yi± ; yi ) are
also trivial in the abelianization and hence we get the same result.
Algebraic & Geometric Topology, Volume 4 (2004)
568
Craig A. Jensen and Nathalie Wahl
References
[1] N. Brady, J. McCammond, J. Meier, and A. Miller, The pure symmetric automorphisms of a free group form a duality group, J. Algebra 246 (2001) 881-896.
[2] K. Brown, Cohomology of Groups, Springer-Verlag, New York-Berlin, 1982.
[3] A. Brownstein and R. Lee, Cohomology of the group of motions of n strings in
3-space, Contemp. Math. vol. 150 (1993) 51-61.
[4] D. J. Collins, Cohomological dimension and symmetric automorphisms of a free
group, Comment. Math. Helv. 64 (1989) 44-61.
[5] M. Culler and K. Vogtmann, Moduli spaces of graphs and automorphisms of free
groups, Invent. Math. 84 (1986) 91-119.
[6] D. Dahm, A generalization of braid theory, Ph.D. Thesis, Princeton Univ., 1962.
[7] D. Goldsmith, The theory of motion groups, Mich. Math. J. 28 (1981), 3-17.
[8] J. L. Harer, Stability of the homology of the mapping class groups of orientable
surfaces, Annals Math. 121 (1985), 215-249.
[9] A. Hatcher, Algebraic Topology, Cambridge University Press, 2002.
[10] A. Hatcher and K. Vogtmann, Cerf theory for graphs, J. London Math. Soc. (2)
58 (1998) 633-655.
[11] A. Hatcher and N. Wahl, Stabilization for the automorphisms of free groups with
boundary, preprint, arXiv:math.GT/0406277.
[12] C. A. Jensen, Cohomology of Aut(Fn ) in the p-rank two case, J. Pure Appl.
Algebra 158 (2001) 41-81.
[13] C. A. Jensen, Contractibility of fixed point sets of auter space, Topology Appl.
119 (2002) 287-304.
[14] C. A. Jensen, J. McCammond and J. Meier, The Euler characteristic of the
Whitehead automorphism group of a free product, preprint.
[15] C. A. Jensen, Homology of holomorphs of free groups, J. Algebra 271 (2004)
281-294.
[16] S. Krstić, Actions of finite groups on graphs and related automorphisms of free
groups, J. Algebra 124 (1989) 119-138.
[17] S. Krstić and K. Vogtmann, Equivariant outer space and automorphisms of
free-by-finite groups, Comment. Math. Helv. 68 (1993) 216-262.
[18] G. Levitt, Automorphisms of hyperbolic groups and graphs of groups, preprint,
arXiv:math.GR/0212088.
[19] I. Madsen and U. Tillmann, The stable mapping class group and Q(CP+∞ ),
Invent. math. 145 (2001), 509-544.
[20] I. Madsen and M. Weiss, The stable moduli space of Riemann surfaces: Mumford’s conjecture, preprint, arXiv:math.AT/0212321.
Algebraic & Geometric Topology, Volume 4 (2004)
Automorphisms of free groups with boundaries
569
[21] W. Magnus, A. Karrass and D. Solitar, Combinatorial group theory. Presentations of groups in terms of generators and relations, Second edition, Dover
Pub., New York, 1976.
[22] J. McCammond and J. Meier, The hypertree poset and the L2 Betti numbers of
the motion group of the trivial link, Math. Ann. 328 (2004) 633-652.
[23] J. McCool, A presentation for the automorphism group of a free group of finite
rank, J. London Math. Soc. (2) 8 (1974),259-266.
[24] J. McCool, Some finitely presented subgroups of the automorphism group of a
free group, J. Algebra 35 (1975), 205-213.
[25] J. McCool, On basis-conjugating automorphisms of the free groups, Can. J.
Math., 38 (1986) no. 6, 1525-1529.
[26] D. McCullough and A. Miller, Symmetric Automorphisms of Free Products,
Mem. Amer. Math. Soc. 122 (1996), no. 582.
[27] D. Quillen, Homotopy properties of the poset of non-trivial p-subgroups of a
group, Advances in Math. 28 (1978) 101-128.
[28] U. Tillmann, On the homotopy of the stable mapping class group, Invent. Math.
130 (1997), 257-275.
[29] N. Wahl, From mapping class groups to automorphisms of free groups, preprint,
arXiv:math.AT/0406278.
Department of Mathematics, University of New Orleans
New Orleans, LA 70148, USA
and
Department of Mathematics, University of Aarhus
Ny Munkegade 116, 8000 Aarhus, DENMARK
Email: jensen@math.uno.edu and wahl@imf.au.dk
Received: 26 February 2004
Revised: 8 July 2004
Algebraic & Geometric Topology, Volume 4 (2004)